There is a need for more effective and efficient predictive data analysis solutions and/or more effective and efficient solutions for generating image representations of categorical/scalar data. Various embodiments of the present invention address one or more of the noted technical challenges. In one example, a method comprises receiving the one or more categorical input features; generating an image representation of the one or more categorical input features, wherein the image representation comprises image region values each associated with a categorical input feature, and further wherein each image region value of the one or more image region values is determined based at least in part on the corresponding categorical input feature associated with the image region value; and processing the image representation using an image-based machine learning model to generate the image-based predictions.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A computer-implemented method for generating an image-based prediction based at least in part on one or more categorical input feature values, the computer-implemented method comprising: receiving the one or more categorical input feature values, wherein each categorical input feature value is associated with a categorical input feature type of one or more categorical input feature types; generating a raw image representation of the one or more categorical input feature values, wherein (i) the raw image representation is associated with one or more raw image region values, (ii) each raw image value is associated with a categorical input feature value of the one or more categorical input feature values, (iii) each raw image region value of the one or more raw image region values is determined based at least in part on the corresponding categorical input feature type associated with the raw image region value, (iv) at least one raw image region value of the one or more image region values is configured to depict a visual representation of textual data associated with the categorical input feature value that is associated with the raw image region value; determining, based at least in part on the raw image representation, one or more raw image region values each associated with a character region of a plurality of character regions within the raw image region; determining, for each character region of the plurality of character regions, a character region scalar value and a character region location within the raw image representation; generating based at least in part on the raw image representation, an image representation of the one or more categorical input feature values to comprise, for each character region of the plurality of character regions, a scalar visual representation of the region scalar value for the character region in the character region location for the character region, wherein (i) the image representation comprises a plurality of pixels, (ii) the image representation is divided into a plurality of image regions each comprising an image region subset of the plurality of pixels, (iii) each image region is associated with an image region value of a plurality of image region values that describes pixel values for the image region pixel subset that is associated with the image region, (iv) each image region of the plurality of image regions is associated with a categorical input feature type of the one or more categorical input feature types, and (v) each image region value is generated in a manner that is configured to represent a categorical input feature value for the corresponding categorical input feature type that is associated with the image region of the image region value; and processing the image representation using an image-based machine learning model to generate an image-based prediction.
This invention relates to a computer-implemented method for generating image-based predictions from categorical input data. The method addresses the challenge of converting non-image categorical data into a visual format that can be processed by image-based machine learning models, which are typically optimized for pixel-based inputs. Categorical data, such as text labels or discrete categories, is often difficult to directly feed into image-based models without transformation. The method begins by receiving one or more categorical input feature values, each associated with a specific categorical feature type. These values are then converted into a raw image representation, where each categorical value is mapped to a visual representation within a designated image region. The raw image includes multiple regions, each corresponding to a categorical feature type, and at least one region visually depicts textual data associated with the categorical value. The method further refines this representation by identifying character regions within the raw image, determining scalar values and locations for each character region, and generating a final image representation. This final image consists of multiple pixels divided into regions, where each region's pixel values encode the categorical input data in a visually interpretable format. The image is then processed by an image-based machine learning model to produce a prediction. This approach enables traditional image-based models to leverage categorical data, improving prediction accuracy in domains where categorical inputs are prevalent.
2. The computer-implemented method of claim 1 , wherein: the one or more categorical input feature values comprise one or more patient features associated with a patient, and the image-based prediction is a health prediction for the patient.
This invention relates to a computer-implemented method for generating health predictions for patients using categorical input feature values. The method involves processing one or more categorical input feature values associated with a patient, such as demographic, medical history, or lifestyle factors, to produce an image-based prediction. The image-based prediction is then used to generate a health prediction for the patient. The method leverages machine learning or statistical models to analyze the categorical input features and derive meaningful insights related to the patient's health status, risk factors, or potential medical conditions. The approach may involve converting categorical data into a format suitable for image-based analysis, such as embedding the features into a visual representation or using image processing techniques to extract patterns. The health prediction can be used for diagnostic purposes, treatment planning, or preventive healthcare strategies. The method aims to improve the accuracy and efficiency of health assessments by integrating structured categorical data with image-based predictive models.
3. The computer-implemented method of claim 1 , wherein the image region value of a plurality of image region values is configured to depict a visual representation of textual data associated with the corresponding categorical input feature type that is associated with the image region of the image region value.
This invention relates to computer-implemented methods for visualizing categorical input feature types in an image. The problem addressed is the need to effectively represent textual data associated with categorical input features in a way that enhances interpretability and usability in image-based systems. The method involves generating an image where specific regions correspond to different categorical input feature types. Each image region is assigned a value that visually represents textual data linked to its associated categorical input feature type. This allows users to quickly identify and interpret the relationship between the categorical data and the image regions. The visual representation may include color coding, patterns, or other graphical elements that make the data more intuitive to understand. The method ensures that the image regions are clearly distinguishable and that the visual representation accurately reflects the underlying textual data. This approach is particularly useful in applications such as data visualization, machine learning model interpretation, and user interface design, where clarity and efficiency in conveying categorical information are critical. By mapping textual data to visual elements in the image, the method improves the accessibility and usability of complex datasets.
4. The computer-implemented method of claim 1 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: determining, for each categorical input feature, a corresponding coordinate grouping of a plurality of coordinate groupings; and generating, for each coordinate grouping of the plurality of coordinate groupings, a coordinate channel; and determining the image representation based on each coordinate channel.
This invention relates to computer-implemented methods for generating image representations from raw image data, particularly for improving the processing and analysis of images in machine learning or computer vision applications. The problem addressed is the efficient and accurate transformation of raw image data into a structured format that preserves relevant categorical features while optimizing computational efficiency. The method involves generating an image representation from a raw image representation by first determining, for each categorical input feature in the raw data, a corresponding coordinate grouping from a set of predefined coordinate groupings. Each coordinate grouping represents a specific spatial or feature-based segmentation of the image. For each of these coordinate groupings, a coordinate channel is generated, which encapsulates the relevant categorical information in a structured format. The final image representation is then derived by combining these coordinate channels, ensuring that the output retains the essential categorical features while being optimized for further processing tasks such as classification, object detection, or segmentation. This approach enhances the accuracy and efficiency of image analysis by systematically organizing categorical data into a structured, channel-based representation.
5. The computer-implemented method of claim 1 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: identifying a plurality of character patterns; generating, for each character pattern of the plurality of character pattern, a feature-based channel of a plurality of feature-based channels, wherein: (i) each feature-based channel comprises one or more feature-based channel region values, and (ii) each feature-based channel region value for a corresponding feature-based channel is associated with the corresponding categorical input feature type, and (iii) each feature-based channel region value for a corresponding feature-based channel is determined based at least in part on whether the corresponding categorical input feature type for the feature-based channel region value comprises the corresponding character pattern associated with the corresponding feature-based channel; and generating the image representation based at least in part on each corresponding feature-based channel of the plurality of coordinate channels.
This invention relates to computer-implemented methods for processing raw image data, particularly for enhancing character recognition in images. The method addresses the challenge of accurately identifying and extracting character patterns from raw image representations, which is crucial for applications like optical character recognition (OCR), document digitization, and automated data entry. The method involves generating an image representation from a raw image by first identifying multiple character patterns within the image. For each identified character pattern, a feature-based channel is generated. Each feature-based channel consists of region values that correspond to specific categorical input feature types, such as edges, corners, or textures. These region values are determined based on whether the categorical input feature type includes the character pattern associated with that channel. The final image representation is then constructed by combining all the feature-based channels, effectively highlighting the relevant character patterns while suppressing irrelevant features. This approach improves the accuracy of character recognition by leveraging feature-based channels to emphasize distinguishing characteristics of each character pattern, making it easier for subsequent processing stages to interpret the image. The method is particularly useful in scenarios where raw image data contains noise or overlapping patterns, as the feature-based channels help isolate and enhance the relevant character information.
6. The computer-implemented method of claim 1 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: generating, based at least in part on the one or more categorical input feature values, one or more coordinate channels and one or more feature-based channels; and merging the one or more coordinate channels and the one or more feature-based channels to generate the image representation.
This invention relates to computer-implemented methods for generating image representations from raw image data, particularly in the context of machine learning or computer vision applications. The problem addressed is the need to efficiently transform raw image data into a structured format that preserves both spatial and feature-based information, enhancing downstream tasks such as classification, segmentation, or object detection. The method involves generating an image representation from a raw image representation by first extracting categorical input feature values from the raw data. These values are used to produce two types of channels: coordinate channels, which encode spatial information, and feature-based channels, which encode non-spatial features derived from the input data. The coordinate channels and feature-based channels are then merged to form the final image representation. This merging process ensures that both spatial and feature-based information are integrated into a unified representation, improving the accuracy and robustness of subsequent machine learning tasks. The approach is particularly useful in applications where raw image data must be processed into a format that balances spatial and feature-based information, such as in medical imaging, autonomous driving, or industrial inspection systems. By generating distinct channels for spatial and feature-based data before merging them, the method avoids information loss and enhances the interpretability of the resulting image representation.
7. The computer-implemented method of claim 1 , wherein: each categorical input feature value of the one or more categorical input feature values is associated with a text formatting pattern, at least one image region value of the one or more image region values is configured to depict a visual representation of textual data associated with the categorical input feature value that is associated with the image region value, and the textual data associated with at least one image region value of the one or more image region values is determined based at least in part on the text formatting pattern for the categorical input feature value that is associated with the image region value.
This invention relates to a computer-implemented method for visualizing categorical data in image-based formats, addressing the challenge of effectively representing textual information derived from categorical input features in a structured and visually coherent manner. The method processes one or more categorical input feature values, each associated with a specific text formatting pattern that defines how the corresponding textual data should be displayed. The system generates one or more image region values, where each image region is configured to depict a visual representation of textual data linked to a categorical input feature value. The textual data for at least one image region is determined based on the predefined text formatting pattern associated with the corresponding categorical input feature value. This ensures consistency and clarity in the visual representation of the data, enhancing readability and interpretability. The method may also involve additional steps such as generating a visual representation of the image region values, where the visual representation includes the textual data formatted according to the associated text formatting patterns. This approach is particularly useful in data visualization applications where categorical data must be presented in a structured and visually appealing format, such as in dashboards, reports, or interactive data displays. The invention improves upon prior art by dynamically linking categorical data to visual representations while maintaining formatting consistency, thereby reducing ambiguity and improving user comprehension.
8. The computer-implemented method of claim 1 , wherein the scalar visual representation for at least one character region of the plurality of character regions is a grayscale visual representation of a character depicted by the character region.
This invention relates to computer-implemented methods for generating scalar visual representations of characters in digital images, particularly for improving character recognition and display. The method addresses the challenge of accurately representing characters in images where visual clarity or contrast may be insufficient for reliable processing. The solution involves converting character regions within an image into scalar visual representations, where each representation is a grayscale depiction of the original character. This grayscale transformation enhances contrast and simplifies subsequent analysis, such as optical character recognition (OCR) or visual display adjustments. The method processes multiple character regions within an image, ensuring that each region is individually converted to its grayscale form. This approach improves the robustness of character recognition systems by standardizing the visual input, reducing variability caused by color or lighting conditions. The grayscale representation also facilitates easier manipulation of character data for further processing or display optimization. The technique is particularly useful in applications requiring high-accuracy text extraction from images, such as document scanning, automated data entry, or image-based text analysis. By converting characters to grayscale, the method ensures consistent and reliable visual input for downstream tasks.
9. The computer-implemented method of claim 1 , wherein the image-based machine learning model comprises a convolutional neural network (CNN).
This invention relates to image-based machine learning systems, specifically those using convolutional neural networks (CNNs) to process and analyze images. The technology addresses the challenge of accurately interpreting visual data in applications such as object recognition, image classification, or automated image analysis, where traditional methods may lack precision or efficiency. The method involves training a CNN to extract features from input images, leveraging its ability to automatically learn hierarchical representations of image data through convolutional layers. The CNN processes the input image by applying a series of convolutional filters, pooling operations, and fully connected layers to generate a feature map or classification output. This approach enhances the model's capability to detect patterns, edges, and higher-level structures within the image, improving accuracy in tasks like identifying objects, detecting anomalies, or segmenting regions of interest. The CNN may include multiple convolutional layers, each applying different filters to progressively refine the feature extraction process. Pooling layers reduce spatial dimensions, retaining important features while reducing computational complexity. The final layers may include fully connected layers that transform the extracted features into a final output, such as a class label or probability distribution. This method is particularly useful in applications requiring real-time image analysis, such as autonomous vehicles, medical imaging, or surveillance systems, where rapid and accurate interpretation of visual data is critical. The use of CNNs ensures robustness and adaptability to diverse image inputs, making the system suitable for deployment in various industries.
10. The computer-implemented method of claim 1 , further comprising automatically scheduling one or more medical visit appointments based at least in part on the image-based prediction.
This invention relates to automated medical appointment scheduling using image-based predictive analytics. The system addresses the challenge of efficiently managing patient visits by leveraging medical imaging data to forecast patient needs and optimize scheduling. The method involves analyzing medical images, such as X-rays, MRIs, or CT scans, to generate predictive insights about a patient's condition. These predictions inform the scheduling of follow-up appointments, ensuring timely care while reducing administrative burden. The system may also consider factors like appointment availability, clinician workload, and patient urgency to determine optimal scheduling. By automating this process, the invention aims to improve healthcare efficiency, reduce wait times, and enhance patient outcomes through data-driven decision-making. The method integrates predictive analytics with existing healthcare systems to streamline workflows and prioritize care based on medical imaging findings.
11. An apparatus for generating an image-based prediction based at least in part on one or more categorical input feature values, the apparatus comprising at least one processor and at least one memory including a computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to: receive the one or more categorical input feature values, wherein each categorical input feature value is associated with a categorical input feature type of one or more categorical input feature types; generate a raw image representation of the one or more categorical input feature values, wherein (i) the raw image representation is associated with one or more raw image region values, (ii) each raw image value is associated with a categorical input feature value of the one or more categorical input feature values, (iii) each raw image region value of the one or more raw image region values is determined based at least in part on the corresponding categorical input feature type associated with the raw image region value, (iv) at least one raw image region value of the one or more image region values is configured to depict a visual representation of textual data associated with the categorical input feature value that is associated with the raw image region value; determine, based at least in part on the raw image representation, one or more raw image region values each associated with a character region of a plurality of character regions within the raw image region; determine, for each character region of the plurality of character regions, a character region scalar value and a character region location within the raw image representation; generate, based at least in part on the raw image representation, an image representation of the one or more categorical input feature values to comprise, for each character region of the plurality of character regions, a scalar visual representation of the region scalar value for the character region in the character region location for the character region, —wherein (i) the image representation comprises a plurality of pixels, (ii) the image representation is divided into a plurality of image regions each comprising an image region subset of the plurality of pixels, (iii) each image region is associated with an image region value of a plurality of image region values that describes pixel values for the image region pixel subset that is associated with the image region, (iv) each image region of the plurality of image regions is associated with a categorical input feature type of the one or more categorical input feature types, and (v) each image region value is generated in a manner that is configured to represent a categorical input feature value for the corresponding categorical input feature type that is associated with the image region of the image region value; and process the image representation using an image-based machine learning model to generate an image-based prediction.
This invention relates to a system for generating image-based predictions from categorical input data. The system addresses the challenge of converting non-image categorical data into a visual format that can be processed by image-based machine learning models. Categorical data, such as text or discrete labels, is often difficult to analyze directly using image-based models, which typically require pixel-based inputs. The apparatus includes at least one processor and memory storing code to perform the following steps. First, it receives one or more categorical input feature values, each associated with a specific categorical feature type. These values are then converted into a raw image representation, where each categorical value is mapped to a visual representation. The raw image is divided into regions, with each region corresponding to a categorical feature type. Some regions may display textual data visually, such as rendering text as pixels. Next, the system identifies character regions within the raw image, determining their scalar values and locations. These regions are then transformed into an optimized image representation, where each character region is visually encoded with its scalar value at the correct location. The final image is structured into multiple regions, each associated with a categorical feature type and containing pixel values that represent the corresponding categorical data. Finally, the processed image is fed into an image-based machine learning model, which generates a prediction based on the visual representation of the categorical input data. This approach enables traditional image-based models to analyze and predict outcomes from non-image categorical data efficiently.
12. The apparatus of claim 11 , wherein: the one or more categorical input feature values comprise one or more patient features associated with a patient, and the image-based prediction is a health prediction for the patient.
13. The apparatus of claim 12 , wherein the image-based machine learning model comprises a convolutional neural network (CNN).
The invention relates to an apparatus for processing images using machine learning, specifically addressing the challenge of efficiently analyzing visual data with high accuracy. The apparatus includes an image-based machine learning model designed to extract features from input images and generate outputs based on those features. The model is implemented as a convolutional neural network (CNN), which is particularly effective for image recognition tasks due to its ability to automatically learn spatial hierarchies of features through convolutional layers. The CNN processes input images by applying convolutional filters to detect patterns such as edges, textures, and shapes, followed by pooling layers to reduce dimensionality while retaining important information. The network may also include fully connected layers for classification or regression tasks. The apparatus may further include preprocessing modules to normalize or enhance input images, as well as post-processing components to refine the model's outputs. The use of a CNN ensures robust performance in tasks like object detection, image classification, and segmentation, making the apparatus suitable for applications in autonomous systems, medical imaging, and quality control. The invention improves upon prior systems by leveraging deep learning techniques to achieve higher accuracy and adaptability in image analysis.
14. The apparatus of claim 11 , wherein the image region value of a plurality of image region values is configured to depict a visual representation of textual data associated with the corresponding categorical input feature type that is associated with the image region of the image region value.
This invention relates to data visualization systems that process categorical input features and generate visual representations in an image. The problem addressed is the need for an effective way to display textual data associated with categorical input features in a structured and visually intuitive manner. The apparatus includes an image generation module that creates an image with multiple image regions, each corresponding to a categorical input feature type. Each image region is assigned an image region value that visually represents textual data linked to its associated categorical input feature type. The image generation module processes the categorical input features and textual data to determine appropriate visual representations, such as color, shape, or pattern, for each image region. The apparatus may also include a data processing module that preprocesses the categorical input features and textual data before visualization. The system ensures that the visual representation in each image region accurately reflects the textual data, enabling users to quickly interpret the information. The invention improves data visualization by providing a clear and organized display of categorical data and its associated textual information.
15. The apparatus of claim 11 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: determining, for each categorical input feature, a corresponding coordinate grouping of a plurality of coordinate groupings; and generating, for each coordinate grouping of the plurality of coordinate groupings, a coordinate channel; and determining the image representation based on each coordinate channel.
This invention relates to image processing systems that generate image representations from raw image data. The problem addressed is the efficient and accurate transformation of raw image data into a structured format suitable for further analysis or machine learning tasks. The apparatus includes a system that processes raw image data by extracting categorical input features and converting them into a structured image representation. The apparatus determines, for each categorical input feature, a corresponding coordinate grouping from a plurality of coordinate groupings. Each coordinate grouping represents a subset of the raw image data associated with a specific categorical feature. The system then generates a coordinate channel for each of these coordinate groupings. A coordinate channel is a structured data format that organizes the raw image data into a format that can be processed by subsequent systems, such as machine learning models. The image representation is then determined based on the combined output of all coordinate channels, resulting in a structured and organized representation of the original raw image data. This approach improves the efficiency and accuracy of image processing by systematically organizing categorical features into a structured format.
16. The apparatus of claim 11 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: identifying a plurality of character patterns; generating, for each character pattern of the plurality of character pattern, a feature-based channel of a plurality of feature-based channels, wherein: (i) each feature-based channel comprises one or more feature-based channel region values, and (ii) each feature-based channel region value for a corresponding feature-based channel is associated with the corresponding categorical input feature type, and (iii) each feature-based channel region value for a corresponding feature-based channel is determined based at least in part on whether the corresponding categorical input feature type for the feature-based channel region value comprises the corresponding character pattern associated with the corresponding feature-based channel; and generating the image representation based at least in part on each corresponding feature-based channel of the plurality of coordinate channels.
This invention relates to image processing, specifically to generating enhanced image representations from raw image data for improved character recognition. The problem addressed is the difficulty in accurately identifying and classifying characters in images, particularly when dealing with variations in font, style, or noise. The solution involves a multi-step process to extract and refine character patterns for better recognition. The apparatus processes raw image data by first identifying multiple character patterns within the image. For each identified character pattern, a feature-based channel is generated. Each feature-based channel contains region values that correspond to specific categorical input feature types, such as shape, stroke, or spatial relationships. These region values are determined by checking whether the categorical input feature type associated with the region includes the character pattern being analyzed. The resulting feature-based channels are then combined to produce a refined image representation that emphasizes relevant character features while suppressing irrelevant or noisy data. This enhanced representation improves the accuracy of subsequent character recognition tasks by providing a more structured and discriminative input. The method is particularly useful in applications like optical character recognition (OCR), document digitization, and automated data extraction.
17. The apparatus of claim 11 , wherein generating the image representation, based at least in part on the raw image representation, further comprises: generating, based at least in part on the one or more categorical input feature values, one or more coordinate channels and one or more feature-based channels; and merging the one or more coordinate channels and the one or more feature-based channels to generate the image representation.
This invention relates to image processing systems that generate enhanced image representations from raw input data. The problem addressed is the need to improve image analysis by incorporating both spatial and categorical information into a unified representation. The apparatus processes raw image data by first extracting one or more categorical input feature values, which describe attributes of the image such as object types, textures, or other non-spatial characteristics. These categorical values are then used to generate coordinate channels, which encode spatial relationships, and feature-based channels, which encode the extracted categorical information. The coordinate and feature-based channels are merged to produce a final image representation that combines spatial and categorical data. This merged representation enhances downstream tasks such as object detection, segmentation, or classification by providing a richer, more informative input. The system is particularly useful in applications where both the location and type of image features are critical, such as medical imaging, autonomous navigation, or industrial inspection. The invention improves upon prior methods by explicitly separating and then integrating spatial and categorical information, leading to more accurate and interpretable image analysis.
18. The apparatus of claim 11 , wherein: each categorical input feature value of the one or more categorical input feature values is associated with a text formatting pattern, at least one image region value of the one or more image region values is configured to depict a visual representation of textual data associated with the categorical input feature value that is associated with the image region value, and the textual data associated with at least one image region value of the one or more image region values is determined based at least in part on the text formatting pattern for the categorical input feature value that is associated with the image region value.
This invention relates to data visualization systems that process categorical input features and image regions to generate visual representations of textual data. The problem addressed is the need for an efficient and intuitive way to display categorical data in a structured visual format, particularly when the data is associated with specific text formatting patterns. The apparatus includes a processing system that receives one or more categorical input feature values, each associated with a distinct text formatting pattern. These patterns define how the corresponding textual data should be visually presented. The system also processes one or more image region values, where each image region is configured to depict a visual representation of textual data linked to a categorical input feature value. The textual data for at least one image region is determined based on the text formatting pattern associated with the corresponding categorical input feature value. This ensures that the visual representation adheres to the predefined formatting rules, enhancing clarity and consistency in data presentation. The apparatus may also include a display system that renders the visual representations of the textual data within the designated image regions, allowing users to interpret the categorical data in a structured and visually coherent manner. The system ensures that the formatting patterns are applied correctly, maintaining uniformity across different data visualizations. This approach improves the usability of data visualization tools, particularly in applications requiring precise and standardized text formatting for categorical data.
19. A non-transitory computer storage medium comprising instructions for generating an image-based prediction based at least in part on one or more categorical input feature values, the instructions being configured to cause one or more processors to at least perform operations configured to: receive the one or more categorical input feature values, wherein each categorical input feature value is associated with a categorical input feature type of one or more categorical input feature types; generate a raw image representation of the one or more categorical input feature values, wherein (i) the raw image representation is associated with one or more raw image region values, (ii) each raw image value is associated with a categorical input feature value of the one or more categorical input feature values, (iii) each raw image region value of the one or more raw image region values is determined based at least in part on the corresponding categorical input feature type associated with the raw image region value, (iv) at least one raw image region value of the one or more image region values is configured to depict a visual representation of textual data associated with the categorical input feature value that is associated with the raw image region value; determine, based at least in part on the raw image representation, one or more raw image region values each associated with a character region of a plurality of character regions within the raw image region; determine, for each character region of the plurality of character regions, a character region scalar value and a character region location within the raw image representation; generate, based at least in part on the raw image representation, an image representation of the one or more categorical input feature values to comprise, for each character region of the plurality of character regions, a scalar visual representation of the region scalar value for the character region in the character region location for the character region, wherein (i) the image representation comprises a plurality of pixels, (ii) the image representation is divided into a plurality of image regions each comprising an image region subset of the plurality of pixels, (iii) each image region is associated with an image region value of a plurality of image region values that describes pixel values for the image region pixel subset that is associated with the image region, (iv) each image region of the plurality of image regions is associated with a categorical input feature type of the one or more categorical input feature types, and (v) each image region value is generated in a manner that is configured to represent a categorical input feature value for the corresponding categorical input feature type that is associated with the image region of the image region value; and process the image representation using an image-based machine learning model to generate an image-based prediction.
This invention relates to a system for generating image-based predictions from categorical input data using machine learning. The problem addressed is the challenge of processing categorical data, which lacks inherent numerical structure, in machine learning models that typically require numerical inputs. The solution involves converting categorical input features into a structured image representation that can be processed by an image-based machine learning model. The system receives one or more categorical input feature values, each associated with a specific categorical feature type. These values are transformed into a raw image representation, where each categorical value is mapped to a specific region within the image. The raw image includes multiple regions, each determined by the corresponding categorical feature type. At least one region visually represents textual data associated with the categorical value. The system then processes the raw image to identify character regions, determining a scalar value and location for each. These regions are refined into an image representation where each character region is visually encoded with its scalar value at the specified location. The final image is divided into multiple regions, each corresponding to a categorical feature type and containing pixel values that represent the associated categorical value. This structured image is then input into an image-based machine learning model, which generates a prediction based on the visual representation of the categorical data. The approach enables traditional image-based models to process and predict from categorical input features effectively.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
October 31, 2019
March 1, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.